CloudTextract

CloudTextract is a self-hosted document data extraction platform built to be run as a business. It reads scanned documents with machine learning rather than plain optical character recognition, so it does not just return a page of characters — it identifies the structure, pulling field-and-value pairs out of forms and rows and columns out of tables. Extraction is powered by Amazon Web Services, on your own AWS account.

The problem it solves is the one most back offices still solve by hand. Data gets copied off PDFs, scans and photographs manually, or through OCR software that needs configuring per template and reconfiguring every time a form changes. Because the model reads the document instead of matching a template, it handles documents it has never seen, and turns work measured in hours or days into minutes.

What it reads

• Text — printed text out of PDFs, PNGs and JPEGs.
• Handwriting — handwritten content, in English.
• Forms — key and value pairs identified as pairs, so "Invoice number" stays attached to the number next to it.
• Tables — table structure preserved rather than flattened into a run of words.
• 6 document languages — English, French, German, Italian, Portuguese and Spanish.

Receipt analysis

Receipts get their own treatment, returning both summary and line-item data. It will find the vendor on a receipt even when the name appears only inside a logo with nothing labelling it as the vendor, and it will pull item, quantity and price out of line items that have no column headers above them — the two things that break template-based extraction on real receipts.

Limits worth knowing before you price a plan

• PNG and JPEG — up to 10 MB per file.
• PDF — up to 500 MB per file and up to 3,000 pages.
• AWS Free Tier — new AWS customers get 3 months covering up to 1,000 pages a month on text tasks and up to 100 pages a month on form or table tasks, which is enough to prove the platform out before spending anything.

The business layer

• Plans — monthly subscriptions and prepaid packs, created from the admin panel.
• 8 payment gateways — PayPal, Stripe, Razorpay, Paystack and Mollie for subscriptions and prepaid; Braintree and Coinbase for prepaid; and offline bank transfer for both. All of them, and every SaaS feature, are covered by the Regular License.
• Crypto — Coinbase takes Bitcoin, Bitcoin Cash, Ethereum, USD Coin, Litecoin, Dogecoin and Dai on prepaid plans.
• Coupons and promo codes — discount codes for prepaid plans.
• Affiliate and referral system — referrals, earnings and payouts, included rather than bolted on.
• Finance dashboard — monthly and yearly income against estimated spend on Textract, which is the pair of figures that tells you whether a plan is priced above cost.
• Projects — group documents per client or case with their own metrics, instead of one flat list of results.
• Admin panel — users, plans, results and settings in one place, including a switch for the upload file preview.
• Auto update — move to a new release without a manual file drop.

Requirements worth knowing

Built on PHP 7.4 and Laravel 8.4, with comprehensive documentation. A valid AWS account is required, since Textract is the engine behind every extraction, and pages are billed by AWS at its own published rates — which is what the estimated-spend dashboard exists to track. The interface is fully responsive.

Ideal for

• Finance and bookkeeping operations turning invoices and receipts into ledger lines without retyping them.
• Expense processing where the useful data is on a photographed receipt with no consistent layout.
• Back offices digitising intake forms, applications and paper archives at volume.
• Logistics and procurement teams reading tables off delivery notes and purchase orders.
• Founders launching a paid document extraction service with plans, gateways and affiliates already built.
• Anyone whose current OCR needs a new template every time a supplier redesigns a form.